IN THIS PAPER, we consider a class of continuous-time recurrent

Transcription

1 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 51, NO. 4, APRIL Global Output Convergence of a Class of Continuous-Time Recurrent Neural Networks With Time-Varying Thresholds Derong Liu, Senior Member, IEEE, Sanqing Hu, Jun Wang, Senior Member, IEEE Abstract This paper discusses the global output convergence of a class of continuous-time recurrent neural networks (RNNs) with globally Lipschitz continuous monotone nondecreasing activation functions locally Lipschitz continuous time-varying thresholds. We establish one sufficient condition to guarantee the global output convergence of this class of neural networks. The present result does not require symmetry in the connection weight matrix. The convergence result is useful in the design of recurrent neural networks with time-varying thresholds. Index Terms Global output convergence, Lipschitz continuity, Lyapunov diagonal semistability, neural networks, time-varying threshold. I. INTRODUCTION IN THIS PAPER, we consider a class of continuous-time recurrent neural networks (RNNs) given by or, equivalently, in matrix format given by is the state vector, is a constant connection weight matrix, is a nonconstant input vector function defined on which is called the time-varying threshold, is a nonlinear vector-valued activation function from to, is called the output of the network (1). When is a constant vector threshold, the RNN model (1) has been applied to content-addressable memory (CAM) problems in [8] [9], is also a subject of study in [1] [16]. Recently, the RNN model (1) has been widely applied in solving various optimization problems such as linear programming problem [18], [19], Manuscript received April 29, 2002; revised January 20, This work was supported by the National Science Foundation under Grant ECS This paper was recommended by Associate Editor A. Kuh. D. Liu S. Hu are with the Department of Electrical Computer Engineering, University of Illinois at Chicago, Chicago, IL USA ( dliu@ieee.org, shu@cil.ece.uic.edu). J. Wang is with the Department of Automation Computer-Aided Engineering, The Chinese University of Hong Kong, Shatin, NT, Hong Kong ( jwang@acae.cuhk.edu.hk). Digital Object Identifier /TCSII (1) shortest path problem [21], sorting problem [20], assignment problem [17], [22]. This class of neural networks has been demonstrated for easy implementation using electronic circuits. The RNN model (1) is different from the well-known Hopfield neural networks which have been used in some optimization problems, e.g., [4], [10], [15]. In some applications of neural networks (e.g., CAM), the convergence of the network in the state space is a basic requirement [13], while in other applications (e.g., some optimization problems), only the convergence in the output space may be required [17], [18], [20] [22]. Recently, global asymptotic stability global exponential stability of the Hopfield neural networks have received attention, e.g., [2] [4], [7], [11], [12], [14], [23]. Within the class of sigmoidal activation functions, it was proved that negative semidefiniteness of the symmetric connection weight matrix of a neural network model is necessary sufficient for absolute stability of the Hopfield neural networks [3]. The absolute stability result was extended to absolute exponential stability in [11]. Within the class of globally Lipschitz continuous monotone nondecreasing activation functions, a series of papers (see, e.g., [4], [12], [23], references cited therein) generalized stability conditions /or conditions on the permitted classes of activation functions as well as the types of stability (absolute, asymptotic, exponential). For the RNN model (1) with constant threshold, which is actually a special case of the general CAM network in [5] (cf. [5, eq. (13)]), a convergence result can easily be obtained that requires symmetry in the connection weight matrix. Convergence of the RNN model (1) with time-varying thresholds has not yet been investigated. There are several reasons for studying RNNs with time-varying thresholds. First, as mentioned in [7], time-varying thresholds can drive quickly to some desired region of activation space. Second, in some RNNs for optimization, it is required that their thresholds vary over time to ensure the feasibility optimality of the solution, as elaborated in [17] [22]. Third, as the thresholds are also adaptive parameters, the convergence issue arises for online learning of the RNNs. Fourth, the thresholds can be considered as external inputs which are usually time varying. Fifth, in a cascade neural network, its inputs are the outputs of the previous layer, which are usually time varying. For the RNN model considered in this paper, due to the time-varying threshold, (1) is a nonautonomous differential equation. Moreover, it does not contain linear terms as those in the Hopfield networks. Hence, the dynamic structure of this class of neural networks is different from that /04$ IEEE

2 162 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 51, NO. 4, APRIL 2004 of the Hopfield models. Since different structures of differential equations usually results in quite different dynamic behaviors, the convergence of (1) is expected to be quite different from those of the Hopfield models. Note that outputs of (1) represent the optimal solutions of the optimization problems in [17], [18], [20] [22], output convergence of (1) is desirable necessary. This paper investigates the global output convergence of the RNN model (1) with globally Lipschitz continuous monotone nondecreasing activation functions locally Lipschitz continuous time-varying thresholds. We establish one sufficient condition for the global output convergence of RNN model (1). The present result extends those in [17], [18], [20] [22] to more general cases of connection weight matrix. As a consequence, the present result exps the application domain of the RNN model (1). The remainder of this paper is organized as follows. In Section II, some preliminaries on recurrent neural networks are presented. Convergence results are developed in Section III. An illustrative example is given in Section IV. Finally, we make concluding remarks in Section V. II. ASSUMPTIONS AND PRELIMINARIES We assume that the function in (1) belongs to the class of globally Lipschitz continuous monotone nondecreasing activation functions; that is, for, there exists such that,,, It should be noted that such activation functions may not be bounded. There are many frequently used activation functions that satisfy this condition, for example,,,,. We assume that the time-varying thresholds are locally Lipschitz continuous satisfy the following conditions: are some constants,, i.e., we assume that Lemma 2.1 (Lemma 1 in [23]): Let be a globally Lipschitz continuous monotone nondecreasing activation function, then For the purpose of our next lemma, we denote may take, respectively. Let (2) (3) Lemma 2.2: If in (3) is not an empty set, there exists at least one constant vector such that Proof: Since in (3) is not an empty set, there exists some,. By the continuity of, there must exist at least one constant As a result, by noting. This completes the proof. Definition 2.1: The RNN model (1) is said to be globally output convergent if, given any, there exists a constant vector It should be noted that in Definition 2.1, may be different from for two different initial condition. Global convergence in the present paper is in the sense that each nonequilibrium solution of (1) converges to an equilibrium state of (1) (cf. [5] [13]). Global (state) convergence is usually shown through the use of an energy function whose value decreases monotonically along nonequilibrium solutions of a neural network. Such results guarantee the nonexistence of nonconstant periodic solutions as well as chaotic solutions in the network. However, due to the squashing effects of the activation function, especially when is given by a hard-limiter type activation function, output convergence of RNN (1) (which is studied in this paper) may not imply state convergence. When RNN (1) is applied to optimization problems, it is the output, not the state, that represents the optimal solutions [17], [18], [20] [22]. It is, therefore, an important issue to analyze the output convergence of RNNs when state convergence analysis is not available (as in the present case). We note that the state convergence analysis of RNN model (1) may be very difficult to achieve due to the use of time-varying threshold. Definition 2.2: An matrix is said to be Lyapunov diagonally semistable if there exists a diagonal matrix with. In the sequel, is the maximum eigenvalue of, is the norm of a vector or a matrix, for a vector, for a matrix. Let III. CONVERGENCE ANALYSIS In this section, we will establish the global output convergence of the RNN model (1). We first prove the following lemma that pertains to the existence uniqueness of solution of the RNN model (1). Lemma 3.1: If (2) is satisfied there exists a constant vector, then, given any, the RNN model (1) has a unique solution defined on. Proof: Let. Then the RNN model (1) can be transformed into the following equivalent system: (4)

3 LIU et al.: GLOBAL OUTPUT CONVERGENCE OF A CLASS OF CONTINUOUS-TIME RNNs 163 is a globally Lipschitz continuous monotone nondecreasing activation function. Let. Since is locally Lipschitz continuous is globally Lipschitz continuous, we see that is locally Lipschitz continuous is globally Lipschitz continuous. As a result, is locally Lipschitz continuous. Based on the wellknown Existence Uniqueness Theorem for solutions of ordinary differential equations [6], for any given initial condition, there is a unique solution to (4) in some time interval or is the maximal right existence interval of the solution. Let be any finite time is a solution of system (4) with any fixed initial point for. In what follows, we will give an estimate of the solution for Since is globally Lipschitz continuous monotone nondecreasing, there exists a positive number,. Therefore with ), If there exists a vector, then the output vector of the RNN model (1) is globally convergent; that is, given any, there exists a vector satisfies may be different from. Proof: Based on the conditions in Theorem 3.1 Remark 3.1, we only need to consider system (4). Let. Then,. We first prove that is bounded for. Since is Lyapunov diagonally semistable, there exists a diagonal matrix with. Consider a function (5) Meanwhile, noticing that is continuous on we have that there exists a positive number,. Hence,. By integrating both sides of (4) from to, we obtain Computing the time derivative of of system (4) yields along the trajectory (6) Thus, we have for all (7). On the other h, in view of (6) Lemma 2.1, we have According to Gronwall s lemma [6], it follows that (8) that is,. Then, from (7) we get Hence, the solution will be bounded for if is finite. Therefore, by virtue of the continuation theorem for the solutions of ordinary differential equations [6], we can conclude that the solution exists in the time interval the solution is unique on. This completes the proof of the lemma. Remark 3.1: According to Lemma 3.1, given any, the RNN model (1) has a unique state solution defined on if (2) is satisfied there exists a constant vector. Moreover, the RNN model (1) can be transformed into the equivalent system (4) is defined on We now establish the main result of the present paper. Theorem 3.1: Suppose that is Lyapunov diagonally semistable (i.e., there exists a diagonal matrix that is Taking integral on both sides yields that is (9)

4 164 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 51, NO. 4, APRIL 2004 which is bounded. As a result, is bounded for according to (8). This implies that,, are all bounded. Now we arbitrarily choose a sequence as. For, three cases are concerned with as follows: i) is bounded. There must exist a subsequence (10) activation function. Similar to the proof of boundedness of, we can see that is bounded, that is, there exists a positive number, Note that is Lyapunov diagonally semistable; that is, there exists a positive definite matrix. Let. Now we consider the following differentiable function (12) ii) may be different from. is unbounded there exists a subsequence. From (9), we have that in (6) is bounded on. Then, one can see that From (8) it follows that (13) is defined in (8). Then is bounded. This implies that. Since is monotone nondecreasing function on, we see that, As a result,,. Let. We have Computing the time derivative of of system (11) yields (14) along the trajectory iii) is unbounded there exists a subsequence. Based on boundedness of in (6), we know that is bounded. This implies that. Since is monotone nondecreasing function on, we see that, As a result,,. Let. We have Based on the three cases above, we can conclude that there exist a subsequence Thus, (15) is given in (13) is a strictly monotone decreasing function on is bounded. As a result, there exists a number (16) Next we will show that. To do so, we consider the following three cases for corresponding to cases i) iii) above, respectively. 1) Noticing the nondecreasing monotonicity of with respect to boundedness of,wehave may be different from. In the following, let. Then the RNN model (1) can be transformed into the following equivalent system that is (11) is a globally Lipschitz continuous monotone nondecreasing 2). Noticing the boundedness of on, similar to ii) above we can

5 LIU et al.: GLOBAL OUTPUT CONVERGENCE OF A CLASS OF CONTINUOUS-TIME RNNs 165 Fig. 1. Output convergence of the network in the example with x =(1; 1;01;01;01). derive that, (i.e., ) Thus, in view of (14) we have 3). Noticing the boundedness of on, similar to iii) above we can derive that, (i.e., ) which leads to. This implies that Based on (11) consequently, ; that is,, it follows that. Since, one can see that. From (1) it follows that In view of 1) 3) above, we have Therefore Then, from (16) it follows that This completes the proof of the theorem. Remark 3.2: In Theorem 3.1, one needs to check the condition that there exists a vector. By Lemma 2.2, this condition is satisfied if in (3) is not an empty set. In Theorem 3.1 the vector satisfying may not be unique when is singular. Remark 3.3: The RNN model (1) with constant thresholds is a special case of the general CAM network (13) in [5]. To guarantee that every trajectory approaches one of equilibrium points, the symmetry condition (15) in [5] is required. However, the weight matrix in Theorem 3.1 may not be symmetric. Remark 3.4: Some specific recurrent neural networks with time-varying thresholds have been studied in [17], [18], [20] [22]. On one h, the time-varying thresholds in [17], [18], [20] [22] satisfy (5). On the other h, one can easily check that the connection weight matrices of the networks

6 166 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 51, NO. 4, APRIL 2004 Fig. 2. Output convergence of the network in the example with x =(040; 40;40;02;03). in [17], [18], [20] [22] satisfy. As a result, Theorem 3.1 extends the results in [17], [18], [20] [22] to more general cases as far as the connection weight matrix is concerned. IV. AN ILLUSTRATIVE EXAMPLE Consider the RNN model (1), It is easy to see that is singular is not symmetric. Letting, one can verify that is negative semidefinite; that is, is Lyapunov diagonally semistable. On the other h, letting the initial condition, Fig. 1 shows that the positive half trajectory of the RNN model (1) converges to the point output of the RNN model (1) converges to the point satisfies with. Since the conditions of Theorem 3.1 are satisfied, this network is globally output convergent. To verify this, we choose a different initial point. Fig. 2 shows that positive half trajectory output of the RNN model (1) all converge the point moreover. Obviously,. In Figs. 1 2, the dotted lines correspond to the solid lines correspond to, V. CONCLUSIONS In this paper, we have established global output convergence for a class of continuous-time recurrent neural networks with globally Lipschitz continuous monotone nondecreasing activation functions locally Lipschitz continuous time-varying thresholds. One sufficient condition to guarantee the global output convergence of this class of neural networks has been established. This result extends existing results is very useful in the design of recurrent neural networks with time-varying thresholds. REFERENCES [1] A. Bhaya, E. Kaszkurewicz, V. S. Kozyakin, Existence stability of a unique equilibrium in continuous-valued discrete-time asynchronous hopfield neural networks, IEEE Trans. Neural Networks, vol. 7, pp , May [2] M. Forti, S. Manetti, M. Marini, A condition for global convergence of a class of symmetric neural circuits, IEEE Trans. Circuits Syst.-I, vol. 39, pp , June [3], Necessary sufficient condition for absolute stability of neural networks, IEEE Trans. Circuits Syst.-I, vol. 41, pp , July [4] M. Forti A. Tesi, New condition for global stability of neural networks with application to linear quadratic programming problems, IEEE Trans. Circuits Syst.-I, vol. 42, pp , July [5] S. Grossberg, Nonlinear neural networks: Principles, mechanism, architectures, Neural Networks, vol. 1, pp , 1988.

7 LIU et al.: GLOBAL OUTPUT CONVERGENCE OF A CLASS OF CONTINUOUS-TIME RNNs 167 [6] J. K. Hale, Ordinary Differential Equations. New York: Wiley, [7] M. W. Hirsch, Convergent activation dynamics in continuous time networks, Neural Networks, vol. 2, pp , [8] J. J. Hopfield, Neuronal networks physical systems with emergent collective computational abilities, Proc. Nat. Acad. Sci., vol. 79, pp , [9], Neurons with graded response have collective computational properties like those of two-state neurons, Proc. Nat. Acad. Sci., vol. 81, pp , [10] J. J. Hopfield D. W. Tank, Neural computation of decisions in optimization problems, Biol. Cybern., vol. 52, pp , [11] X. B. Liang T. Yamaguchi, Necessary sufficient condition for absolute exponential stability of Hopfield-type neural networks, IEICE Trans. Inform. Syst, vol. E79-D, pp , July [12] X. B. Liang J. Si, Global exponential stability of neural networks with globally Lipschitz continuous activations its application to linear variational inequality problem, IEEE Trans. Neural Networks, vol. 12, pp , Mar [13] A. N. Michel D. Liu, Qualitative Analysis Synthesis of Recurrent Neural Networks. New York: Marcel Dekker, [14] H. Qiao, J. G. Peng, Z. B. Xu, Nonlinear measures: A new approach to exponential stability analysis for Hopfield-type neural networks, IEEE Trans. Neural Networks, vol. 12, pp , Mar [15] D. W. Tank J. J. Hopfield, Simple neural optimization networks: An A/D converter, signal decision circuit, a linear programming circuit, IEEE Trans. Circuits Syst., vol. 33, pp , May [16] M. Takeda J. W. Goodman, Neural networks for computation: Number representations programming complexity, Appl. Opt., vol. 25, pp , [17] J. Wang, Analog neural networks for solving the assignment problem, Electron. Lett., vol. 28, pp , Nov [18], Analysis design of a recurrent neural network for linear programming, IEEE Trans. Circuits Syst.-I, vol. 40, pp , Sept [19], A deterministic annealing neural network for convex programming, Neural Networks, vol. 7, pp , [20], Analysis design of an analog sorting network, IEEE Trans. Neural Networks, vol. 6, pp , July [21], A recurrent neural network for solving the shortest path problem, IEEE Trans. Circuits Syst.-I, vol. 43, pp , June [22], Primal dual assignment networks, IEEE Trans. Neural Networks, vol. 8, pp , May [23] Y. Zhang, P. A. Heng, A. W. C. Fu, Estimate of exponential convergence rate exponential stability for neural networks, IEEE Trans. Neural Networks, vol. 10, pp , Nov